Alcohol Availability and Public Health

Causal Evidence from a Post-Rationing Reform

Tomas Reivinger

04 June, 2025

1 TODO

  • Remove section numbers completley?
  • The images before the table or NOT in svg format and therefore blurry. fix.
  • Read the tables and write in significance stars with red ink in case I get asked about it!

2 Motivation

It’s a big problem

Limited credible causal evidence

2.1 Big problem

…

2.2 Little causal evidence

  • A lot of evidence for a detrimental effect of alcohol on health

  • Difficult to identify causal …

  • Natural experiments, a fix?

3 Roadmap

  1. Motivation
  2. Historical Background
  3. Institutional setting
  4. Data
  5. Empirical strategy and design
  6. Results
  7. Robustness
  8. Discussion

4 Historical Background

  • Before Bratt: high levels of alcohol consumption and the 1922 prohibition referendum
  • The Bratt system: a system of individual control
  • After Bratt: Systembolaget

5 Institutional setting

  • TODO: Edit these 2 in oblsidian and remove whitespace betwen them and save as SVG

5.1 A snapshot from 1955

  • No delivery-point sales on the municipality \(\times\) year level

6 Data

  • All cause mortality (1968-2023)

  • Motor vehicle accidents (1985-2023)

  • Sales in Liters of alcohol (1978-2008)

6.1 Municipality reforms and sample restrictions

  • 900 municipalities in 1968, 288 in 1996 (Source: REGINA, SCB)

  • Mortality data aggregated to 1995 municipal borders

INSERT DCDH AND CS PLOT HERE (MAKE A NEW ONE!) + Or maybe skip as this reveals the result? Maybe add at the end, after robustness?

7 Empirical strategy and design

The population equation of interest is

\[\begin{equation} Y_{mt} = \gamma_m + \lambda_t + \beta_{mt} \text{store}_{mt} + \epsilon_{mt}\, , \label{eq:pop-eq-of-interest} \end{equation}\]

  • Callaway and Sant’Anna (2021)
  • Long differences with base-period \(t=-1\)
  • Time-invariant controls (population size)

8 Results

  1. Sales
  2. Mortality
  3. Motor vehicle accidents

8.1 Staggered adoption schedule (sales)

  • 1st store test uses municipalities with no store a the control group

  • 2nd store test uses municipalities with only one store as control group

8.2 First and second store results

The 1st store test (mechanically) confirms that the treatment is well-defined, i.e. municipalities see positive in-store sales compared to municipalities without a store.

  • (The reason why the first coefficient is much lower is because of late in the year store openings)

The 2nd store test is more interesting. It shows that municipalities that open a 2nd store increase their sales with about 25%.

8.3 Staggered adoption schedule (mortality)

  • temp

8.4 All-cause mortality

  • Decide if you want one or 2 columns (depends on if you want any text. IWait until manuscript is done.)

8.5 Population size

  • temp

8.6 All-cause mortality adjusting for population size

  • tempxxxxx

8.7 MVs

  • temp

9 Dose size compared

Caveats:

  • Substitution effect?
    • Moonshine
    • Pickup points (add percentages here)
    • Other sources of alcohol

If we multiply the coefficient on total alcohol sales from the ‘Saturday’ experiment by the number of days Systembolaget is open (today), we get \(0.035 \times 6 = 0.21\), compared to the \(0.25\) from the 2nd store test (column 5).

10 Robustness

…

10.1 SUTVA

Describe the test and treatment df